GameCult/CultLib
Watch
9
0
Fork
You've already forked CultLib
 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

Maps: record the C2a rollout-order and mark-by-content rulings, publication Cuts 1-2 landing

Browse source 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Meta Vens 2026-09-29 19:34:54 +02:00
commit 5adbcaa0c8
2 changed files with 33 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

18
docs/cultcache-publication-cut.md
View file

@ -1,7 +1,21 @@
# CultCache publication: the cache owns "published once", and order stays data
Status: cut map, Imagination pass 1, 2026-09-29. It is not committed yet. Self
commits it to `main` as `docs/cultcache-publication-cut.md`. The ends are the
Status, 2026-09-30: Cuts 1-2 landed on `main` at `2a1b8a8`, from `hands/cultcache-pub-c1c2` at `5fb3fca`.
Soul verdict: merge. Suites green: Caching 323, Networking 295, Mesh 262. Cut 3 is in Hands on
`hands/cultcache-pub-c3`, with Soul's Cuts 1-2 follow-ups folded in:
- F4: pin `Held`'s publish-on-throw guard. This is needed before Cut 4 relies on sequence density; Stryker could
not mutate `Held` (CS0165).
- F3: pin observer removal on unsubscribe.
- F1: the "getters run once" claim is false for conditional commits on variant stores.
Recorded:
- F2, the Q-P2 hazard. A rethrown observer exception makes a committed write look cancelled or refused. The
only live typed catch around a cache write is AetheriaEve `Aetheria.State.Daemon/Program.cs:3722`. It needs a
fail-fast handler, and no production code installs one.
- F5: subscribe and unsubscribe cost O(n).
- F6: writes after `Dispose` are dropped silently.
Map written by Imagination pass 1, 2026-09-29. The ends are the
operator ruling of 2026-09-30: every committed cache change reaches subscribers
exactly once. This document owns the means.

17
docs/document-variants-cut.md
View file

@ -339,6 +339,23 @@ Soul held it. Fixes are in Hands:
- **C2b precondition (Soul F7).** Whether a pre-id element keeps its load-time id depends on the write path.
C2b must refuse an override that targets a record whose ids exist only in memory, or mint on open.
**C2a Soul pass 2 (`e85b201b`), 2026-09-30: hold.** The tripwire only works against unreleased readers.
Every published release (`cultlib-unity-v1.0.60`, `caching-unity-v1.4.0`, `cultcache-ts-v0.14.0`,
`cultcache-py-v0.3.0`) predates C0. The 1.0.60 C# reader loads a v3 store, and its next write is v1 with the ids
stripped (probe). Other findings:
- F2: the same element object twice in one list commits, then the store refuses to load.
- F3: CultMesh single-file writes bypass the marker.
- F4: TS `pushAll` writes the header without reading it first.
**C2a rulings, 2026-09-30 (operator):**
- **Rollout order closes the released-reader hole.** Ship a CultLib release that refuses v3, move every C#
reader of a store onto it, then let a C2a writer touch that store. There is no format change for this.
This supersedes the claim above that old readers are tripwired: that claim holds only for post-C0 readers.
- **Mark by content.** A store is written as v3 (v5 for directory stores) only when a record actually holds an
element id. An existing v3 header on disk stays.
- Fix batch 3 in Hands: F2, marking by content in every runtime, F3 routed through the cache's header
decision and load checks, F4.
**Operator rulings, 2026-09-30:**
- **Duplicate unique-index values do not load** ("Duplicate indices should not load"). This covers
plain records too, so R6-for-indexes becomes one general rule: